/*
** The unixFile structure is subclass of sqlite3_file specific to the unix
** VFS implementations.
*/
typedef struct unixFile unixFile;
struct unixFile {
  sqlite3_io_methods const *pMethod;  /* Always the first entry */
  struct unixOpenCnt *pOpen;       /* Info about all open fd's on this inode */
  struct unixLockInfo *pLock;      /* Info about locks on this inode */
  int h;                           /* The file descriptor */
  int dirfd;                       /* File descriptor for the directory */
  unsigned char eFileLock;          /* The type of lock held on this fd */
  int lastErrno;                   /* The unix errno from the last I/O error */
  void *lockingContext;            /* Locking style specific state */
  UnixUnusedFd *pUnused;           /* Pre-allocated UnixUnusedFd */
  int fileFlags;                   /* Miscellanous flags */
  const char *zPath;               /* Name of the file */
  unixShm *pShm;                   /* Shared memory segment information */
#if SQLITE_ENABLE_LOCKING_STYLE
  int openFlags;                   /* The flags specified at open() */
#endif
#if SQLITE_ENABLE_LOCKING_STYLE || defined(__APPLE__)
  unsigned fsFlags;                 /* cached details from statfs() */
#endif
#if SQLITE_THREADSAFE && defined(__linux__)
  pthread_t tid;                   /* The thread that "owns" this unixFile */
#endif
#if OS_VXWORKS
  int isDelete;                    /* Delete on close if true */
  struct vxworksFileId *pId;       /* Unique file ID */
#endif
#ifndef NDEBUG
  /* The next group of variables are used to track whether or not the
  ** transaction counter in bytes 24-27 of database files are updated
  ** whenever any part of the database changes.  An assertion fault will
  ** occur if a file is updated without also updating the transaction
  ** counter.  This test is made to avoid new problems similar to the
................................................................................
#else
#define threadid 0
#endif


/*
** Helper functions to obtain and relinquish the global mutex. The
** global mutex is used to protect the unixOpenCnt, unixLockInfo and
** vxworksFileId objects used by this file, all of which may be 
** shared by multiple threads.
**
** Function unixMutexHeld() is used to assert() that the global mutex 
** is held when required. This function is only used as part of assert() 
** statements. e.g.
**
................................................................................
** POSIX lock if the internal lock structure transitions between
** a locked and an unlocked state.
**
** But wait:  there are yet more problems with POSIX advisory locks.
**
** If you close a file descriptor that points to a file that has locks,
** all locks on that file that are owned by the current process are
** released.  To work around this problem, each unixFile structure contains
** a pointer to an unixOpenCnt structure.  There is one unixOpenCnt structure
** per open inode, which means that multiple unixFile can point to a single
** unixOpenCnt.  When an attempt is made to close an unixFile, if there are
** other unixFile open on the same inode that are holding locks, the call
** to close() the file descriptor is deferred until all of the locks clear.
** The unixOpenCnt structure keeps a list of file descriptors that need to
** be closed and that list is walked (and cleared) when the last lock
** clears.
**
** Yet another problem:  LinuxThreads do not play well with posix locks.
**
** Many older versions of linux use the LinuxThreads library which is
** not posix compliant.  Under LinuxThreads, a lock created by thread
................................................................................
** A cannot be modified or overridden by a different thread B.
** Only thread A can modify the lock.  Locking behavior is correct
** if the appliation uses the newer Native Posix Thread Library (NPTL)
** on linux - with NPTL a lock created by thread A can override locks
** in thread B.  But there is no way to know at compile-time which
** threading library is being used.  So there is no way to know at
** compile-time whether or not thread A can override locks on thread B.
** We have to do a run-time check to discover the behavior of the
** current process.
**
** On systems where thread A is unable to modify locks created by
** thread B, we have to keep track of which thread created each
** lock.  Hence there is an extra field in the key to the unixLockInfo
** structure to record this information.  And on those systems it
** is illegal to begin a transaction in one thread and finish it
** in another.  For this latter restriction, there is no work-around.
** It is a limitation of LinuxThreads.
*/

/*
** Set or check the unixFile.tid field.  This field is set when an unixFile
** is first opened.  All subsequent uses of the unixFile verify that the
** same thread is operating on the unixFile.  Some operating systems do
** not allow locks to be overridden by other threads and that restriction
** means that sqlite3* database handles cannot be moved from one thread
** to another while locks are held.
**
** Version 3.3.1 (2006-01-15):  unixFile can be moved from one thread to
** another as long as we are running on a system that supports threads
** overriding each others locks (which is now the most common behavior)
** or if no locks are held.  But the unixFile.pLock field needs to be
** recomputed because its key includes the thread-id.  See the 
** transferOwnership() function below for additional information
*/
#if SQLITE_THREADSAFE && defined(__linux__)
# define SET_THREADID(X)   (X)->tid = pthread_self()
# define CHECK_THREADID(X) (threadsOverrideEachOthersLocks==0 && \
                            !pthread_equal((X)->tid, pthread_self()))
#else
# define SET_THREADID(X)
# define CHECK_THREADID(X) 0
#endif

/*
** An instance of the following structure serves as the key used
** to locate a particular unixOpenCnt structure given its inode.  This
** is the same as the unixLockKey except that the thread ID is omitted.
*/
struct unixFileId {
  dev_t dev;                  /* Device number */
#if OS_VXWORKS
  struct vxworksFileId *pId;  /* Unique file ID for vxworks. */
#else
  ino_t ino;                  /* Inode number */
#endif
};

/*
** An instance of the following structure serves as the key used
** to locate a particular unixLockInfo structure given its inode.
**
** If threads cannot override each others locks (LinuxThreads), then we
** set the unixLockKey.tid field to the thread ID.  If threads can override
** each others locks (Posix and NPTL) then tid is always set to zero.
** tid is omitted if we compile without threading support or on an OS
** other than linux.
*/
struct unixLockKey {
  struct unixFileId fid;  /* Unique identifier for the file */
#if SQLITE_THREADSAFE && defined(__linux__)
  pthread_t tid;  /* Thread ID of lock owner. Zero if not using LinuxThreads */
#endif
};

/*
** An instance of the following structure is allocated for each open
** inode.  Or, on LinuxThreads, there is one of these structures for
** each inode opened by each thread.
**
** A single inode can have multiple file descriptors, so each unixFile
** structure contains a pointer to an instance of this object and this
** object keeps a count of the number of unixFile pointing to it.
*/
struct unixLockInfo {
  struct unixLockKey lockKey;     /* The lookup key */
  int nShared;                    /* Number of SHARED locks held */
  int eFileLock;                   /* One of SHARED_LOCK, RESERVED_LOCK etc. */
  int nRef;                       /* Number of pointers to this structure */
#if defined(SQLITE_ENABLE_LOCKING_STYLE)
  unsigned long long sharedByte;  /* for AFP simulated shared lock */
#endif
  struct unixLockInfo *pNext;     /* List of all unixLockInfo objects */
  struct unixLockInfo *pPrev;     /*    .... doubly linked */
};

/*
** An instance of the following structure is allocated for each open
** inode.  This structure keeps track of the number of locks on that
** inode.  If a close is attempted against an inode that is holding
** locks, the close is deferred until all locks clear by adding the
** file descriptor to be closed to the pending list.
**
** TODO:  Consider changing this so that there is only a single file
** descriptor for each open file, even when it is opened multiple times.
** The close() system call would only occur when the last database
** using the file closes.
*/
struct unixOpenCnt {
  struct unixFileId fileId;   /* The lookup key */
  int nRef;                   /* Number of pointers to this structure */
  int nLock;                  /* Number of outstanding locks */
  UnixUnusedFd *pUnused;      /* Unused file descriptors to close */
#if OS_VXWORKS
  sem_t *pSem;                     /* Named POSIX semaphore */
  char aSemName[MAX_PATHNAME+2];   /* Name of that semaphore */
#endif
  struct unixOpenCnt *pNext, *pPrev;   /* List of all unixOpenCnt objects */


};

/*
** Lists of all unixLockInfo and unixOpenCnt objects.  These used to be hash
** tables.  But the number of objects is rarely more than a dozen and
** never exceeds a few thousand.  And lookup is not on a critical
** path so a simple linked list will suffice.

*/
static struct unixLockInfo *lockList = 0;
static struct unixOpenCnt *openList = 0;

/*
** This variable remembers whether or not threads can override each others
** locks.
**
**    0:  No.  Threads cannot override each others locks.  (LinuxThreads)
**    1:  Yes.  Threads can override each others locks.  (Posix & NLPT)
**   -1:  We don't know yet.
**
** On some systems, we know at compile-time if threads can override each
** others locks.  On those systems, the SQLITE_THREAD_OVERRIDE_LOCK macro
** will be set appropriately.  On other systems, we have to check at
** runtime.  On these latter systems, SQLTIE_THREAD_OVERRIDE_LOCK is
** undefined.
**
** This variable normally has file scope only.  But during testing, we make
** it a global so that the test code can change its value in order to verify
** that the right stuff happens in either case.
*/
#if SQLITE_THREADSAFE && defined(__linux__)
#  ifndef SQLITE_THREAD_OVERRIDE_LOCK
#    define SQLITE_THREAD_OVERRIDE_LOCK -1
#  endif
#  ifdef SQLITE_TEST
int threadsOverrideEachOthersLocks = SQLITE_THREAD_OVERRIDE_LOCK;
#  else
static int threadsOverrideEachOthersLocks = SQLITE_THREAD_OVERRIDE_LOCK;
#  endif
#endif

/*
** This structure holds information passed into individual test
** threads by the testThreadLockingBehavior() routine.
*/
struct threadTestData {
  int fd;                /* File to be locked */
  struct flock lock;     /* The locking operation */
  int result;            /* Result of the locking operation */
};

#if SQLITE_THREADSAFE && defined(__linux__)
/*
** This function is used as the main routine for a thread launched by
** testThreadLockingBehavior(). It tests whether the shared-lock obtained
** by the main thread in testThreadLockingBehavior() conflicts with a
** hypothetical write-lock obtained by this thread on the same file.
**
** The write-lock is not actually acquired, as this is not possible if 
** the file is open in read-only mode (see ticket #3472).
*/ 
static void *threadLockingTest(void *pArg){
  struct threadTestData *pData = (struct threadTestData*)pArg;
  pData->result = fcntl(pData->fd, F_GETLK, &pData->lock);
  return pArg;
}
#endif /* SQLITE_THREADSAFE && defined(__linux__) */


#if SQLITE_THREADSAFE && defined(__linux__)
/*
** This procedure attempts to determine whether or not threads
** can override each others locks then sets the 
** threadsOverrideEachOthersLocks variable appropriately.
*/
static void testThreadLockingBehavior(int fd_orig){
  int fd;
  int rc;
  struct threadTestData d;
  struct flock l;
  pthread_t t;

  fd = dup(fd_orig);
  if( fd<0 ) return;
  memset(&l, 0, sizeof(l));
  l.l_type = F_RDLCK;
  l.l_len = 1;
  l.l_start = 0;
  l.l_whence = SEEK_SET;
  rc = fcntl(fd_orig, F_SETLK, &l);
  if( rc!=0 ) return;
  memset(&d, 0, sizeof(d));
  d.fd = fd;
  d.lock = l;
  d.lock.l_type = F_WRLCK;
  if( pthread_create(&t, 0, threadLockingTest, &d)==0 ){
    pthread_join(t, 0);
  }
  close(fd);
  if( d.result!=0 ) return;
  threadsOverrideEachOthersLocks = (d.lock.l_type==F_UNLCK);
}
#endif /* SQLITE_THREADSAFE && defined(__linux__) */

/*
** Release a unixLockInfo structure previously allocated by findLockInfo().
**
** The mutex entered using the unixEnterMutex() function must be held
** when this function is called.
*/
static void releaseLockInfo(struct unixLockInfo *pLock){

  assert( unixMutexHeld() );
  if( pLock ){
    pLock->nRef--;
    if( pLock->nRef==0 ){
      if( pLock->pPrev ){
        assert( pLock->pPrev->pNext==pLock );
        pLock->pPrev->pNext = pLock->pNext;
      }else{
        assert( lockList==pLock );
        lockList = pLock->pNext;


      }
      if( pLock->pNext ){
        assert( pLock->pNext->pPrev==pLock );
        pLock->pNext->pPrev = pLock->pPrev;
      }
      sqlite3_free(pLock);
    }
  }
}

/*
** Release a unixOpenCnt structure previously allocated by findLockInfo().
**
** The mutex entered using the unixEnterMutex() function must be held
** when this function is called.
*/
static void releaseOpenCnt(struct unixOpenCnt *pOpen){
  assert( unixMutexHeld() );
  if( pOpen ){
    pOpen->nRef--;
    if( pOpen->nRef==0 ){
      if( pOpen->pPrev ){
        assert( pOpen->pPrev->pNext==pOpen );
        pOpen->pPrev->pNext = pOpen->pNext;
      }else{
        assert( openList==pOpen );
        openList = pOpen->pNext;
      }
      if( pOpen->pNext ){
        assert( pOpen->pNext->pPrev==pOpen );
        pOpen->pNext->pPrev = pOpen->pPrev;
      }
#if SQLITE_THREADSAFE && defined(__linux__)
      assert( !pOpen->pUnused || threadsOverrideEachOthersLocks==0 );
#endif

      /* If pOpen->pUnused is not null, then memory and file-descriptors
      ** are leaked.
      **
      ** This will only happen if, under Linuxthreads, the user has opened
      ** a transaction in one thread, then attempts to close the database
      ** handle from another thread (without first unlocking the db file).
      ** This is a misuse.  */
      sqlite3_free(pOpen);
    }
  }
}

/*
** Given a file descriptor, locate unixLockInfo and unixOpenCnt structures that
** describes that file descriptor.  Create new ones if necessary.  The
** return values might be uninitialized if an error occurs.
**
** The mutex entered using the unixEnterMutex() function must be held
** when this function is called.
**
** Return an appropriate error code.
*/
static int findLockInfo(
  unixFile *pFile,               /* Unix file with file desc used in the key */
  struct unixLockInfo **ppLock,  /* Return the unixLockInfo structure here */
  struct unixOpenCnt **ppOpen    /* Return the unixOpenCnt structure here */

){
  int rc;                        /* System call return code */
  int fd;                        /* The file descriptor for pFile */
  struct unixLockKey lockKey;    /* Lookup key for the unixLockInfo structure */
  struct unixFileId fileId;      /* Lookup key for the unixOpenCnt struct */
  struct stat statbuf;           /* Low-level file information */
  struct unixLockInfo *pLock = 0;/* Candidate unixLockInfo object */
  struct unixOpenCnt *pOpen;     /* Candidate unixOpenCnt object */


  assert( unixMutexHeld() );

  /* Get low-level information about the file that we can used to
  ** create a unique name for the file.
  */
  fd = pFile->h;
................................................................................
    if( rc!=0 ){
      pFile->lastErrno = errno;
      return SQLITE_IOERR;
    }
  }
#endif

  memset(&lockKey, 0, sizeof(lockKey));
  lockKey.fid.dev = statbuf.st_dev;
#if OS_VXWORKS
  lockKey.fid.pId = pFile->pId;
#else
  lockKey.fid.ino = statbuf.st_ino;
#endif
#if SQLITE_THREADSAFE && defined(__linux__)
  if( threadsOverrideEachOthersLocks<0 ){
    testThreadLockingBehavior(fd);
  }
  lockKey.tid = threadsOverrideEachOthersLocks ? 0 : pthread_self();
#endif
  fileId = lockKey.fid;
  if( ppLock!=0 ){
    pLock = lockList;
    while( pLock && memcmp(&lockKey, &pLock->lockKey, sizeof(lockKey)) ){
      pLock = pLock->pNext;
    }
    if( pLock==0 ){
      pLock = sqlite3_malloc( sizeof(*pLock) );
      if( pLock==0 ){
        rc = SQLITE_NOMEM;
        goto exit_findlockinfo;
      }
      memcpy(&pLock->lockKey,&lockKey,sizeof(lockKey));
      pLock->nRef = 1;
      pLock->nShared = 0;
      pLock->eFileLock = 0;
#if defined(SQLITE_ENABLE_LOCKING_STYLE)
      pLock->sharedByte = 0;
#endif
      pLock->pNext = lockList;
      pLock->pPrev = 0;
      if( lockList ) lockList->pPrev = pLock;
      lockList = pLock;
    }else{
      pLock->nRef++;
    }
    *ppLock = pLock;
  }
  if( ppOpen!=0 ){
    pOpen = openList;
    while( pOpen && memcmp(&fileId, &pOpen->fileId, sizeof(fileId)) ){
      pOpen = pOpen->pNext;
    }
    if( pOpen==0 ){

      pOpen = sqlite3_malloc( sizeof(*pOpen) );
      if( pOpen==0 ){
        releaseLockInfo(pLock);

        rc = SQLITE_NOMEM;
        goto exit_findlockinfo;
      }
      memset(pOpen, 0, sizeof(*pOpen));
      pOpen->fileId = fileId;

      pOpen->nRef = 1;
      pOpen->pNext = openList;
      if( openList ) openList->pPrev = pOpen;


      openList = pOpen;
    }else{
      pOpen->nRef++;
    }
    *ppOpen = pOpen;
  }

exit_findlockinfo:
  return rc;
}

/*
** If we are currently in a different thread than the thread that the
** unixFile argument belongs to, then transfer ownership of the unixFile
** over to the current thread.
**
** A unixFile is only owned by a thread on systems that use LinuxThreads.
**
** Ownership transfer is only allowed if the unixFile is currently unlocked.
** If the unixFile is locked and an ownership is wrong, then return
** SQLITE_MISUSE.  SQLITE_OK is returned if everything works.
*/
#if SQLITE_THREADSAFE && defined(__linux__)
static int transferOwnership(unixFile *pFile){
  int rc;
  pthread_t hSelf;
  if( threadsOverrideEachOthersLocks ){
    /* Ownership transfers not needed on this system */

    return SQLITE_OK;
  }
  hSelf = pthread_self();
  if( pthread_equal(pFile->tid, hSelf) ){
    /* We are still in the same thread */
    OSTRACE(("No-transfer, same thread\n"));
    return SQLITE_OK;
  }
  if( pFile->eFileLock!=NO_LOCK ){
    /* We cannot change ownership while we are holding a lock! */
    return SQLITE_MISUSE_BKPT;
  }
  OSTRACE(("Transfer ownership of %d from %d to %d\n",
            pFile->h, pFile->tid, hSelf));
  pFile->tid = hSelf;
  if (pFile->pLock != NULL) {
    releaseLockInfo(pFile->pLock);
    rc = findLockInfo(pFile, &pFile->pLock, 0);
    OSTRACE(("LOCK    %d is now %s(%s,%d)\n", pFile->h,
           azFileLock(pFile->eFileLock),
           azFileLock(pFile->pLock->eFileLock), pFile->pLock->nShared));
    return rc;
  } else {
    return SQLITE_OK;
  }
}
#else  /* if not SQLITE_THREADSAFE */
  /* On single-threaded builds, ownership transfer is a no-op */
# define transferOwnership(X) SQLITE_OK
#endif /* SQLITE_THREADSAFE */


/*
** This routine checks if there is a RESERVED lock held on the specified
** file by this or any other process. If such a lock is held, set *pResOut
** to a non-zero value otherwise *pResOut is set to zero.  The return value
** is set to SQLITE_OK unless an I/O error occurs during lock checking.
................................................................................
  int rc = SQLITE_OK;
  int reserved = 0;
  unixFile *pFile = (unixFile*)id;

  SimulateIOError( return SQLITE_IOERR_CHECKRESERVEDLOCK; );

  assert( pFile );
  unixEnterMutex(); /* Because pFile->pLock is shared across threads */

  /* Check if a thread in this process holds such a lock */
  if( pFile->pLock->eFileLock>SHARED_LOCK ){
    reserved = 1;
  }

  /* Otherwise see if some other process holds it.
  */
#ifndef __DJGPP__
  if( !reserved ){
................................................................................
  ** The reason a single byte cannot be used instead of the 'shared byte
  ** range' is that some versions of windows do not support read-locks. By
  ** locking a random byte from a range, concurrent SHARED locks may exist
  ** even if the locking primitive used is always a write-lock.
  */
  int rc = SQLITE_OK;
  unixFile *pFile = (unixFile*)id;
  struct unixLockInfo *pLock = pFile->pLock;
  struct flock lock;
  int s = 0;
  int tErrno = 0;

  assert( pFile );
  OSTRACE(("LOCK    %d %s was %s(%s,%d) pid=%d (unix)\n", pFile->h,
      azFileLock(eFileLock), azFileLock(pFile->eFileLock),
      azFileLock(pLock->eFileLock), pLock->nShared , getpid()));

  /* If there is already a lock of this type or more restrictive on the
  ** unixFile, do nothing. Don't use the end_lock: exit path, as
  ** unixEnterMutex() hasn't been called yet.
  */
  if( pFile->eFileLock>=eFileLock ){
    OSTRACE(("LOCK    %d %s ok (already held) (unix)\n", pFile->h,
................................................................................
  **  (2) SQLite never explicitly requests a pendig lock.
  **  (3) A shared lock is always held when a reserve lock is requested.
  */
  assert( pFile->eFileLock!=NO_LOCK || eFileLock==SHARED_LOCK );
  assert( eFileLock!=PENDING_LOCK );
  assert( eFileLock!=RESERVED_LOCK || pFile->eFileLock==SHARED_LOCK );

  /* This mutex is needed because pFile->pLock is shared across threads
  */
  unixEnterMutex();

  /* Make sure the current thread owns the pFile.
  */
  rc = transferOwnership(pFile);
  if( rc!=SQLITE_OK ){
    unixLeaveMutex();
    return rc;
  }
  pLock = pFile->pLock;

  /* If some thread using this PID has a lock via a different unixFile*
  ** handle that precludes the requested lock, return BUSY.
  */
  if( (pFile->eFileLock!=pLock->eFileLock && 
          (pLock->eFileLock>=PENDING_LOCK || eFileLock>SHARED_LOCK))
  ){
    rc = SQLITE_BUSY;
    goto end_lock;
  }

  /* If a SHARED lock is requested, and some thread using this PID already
  ** has a SHARED or RESERVED lock, then increment reference counts and
  ** return SQLITE_OK.
  */
  if( eFileLock==SHARED_LOCK && 
      (pLock->eFileLock==SHARED_LOCK || pLock->eFileLock==RESERVED_LOCK) ){
    assert( eFileLock==SHARED_LOCK );
    assert( pFile->eFileLock==0 );
    assert( pLock->nShared>0 );
    pFile->eFileLock = SHARED_LOCK;
    pLock->nShared++;
    pFile->pOpen->nLock++;
    goto end_lock;
  }


  /* A PENDING lock is needed before acquiring a SHARED lock and before
  ** acquiring an EXCLUSIVE lock.  For the SHARED lock, the PENDING will
  ** be released.
................................................................................
  }


  /* If control gets to this point, then actually go ahead and make
  ** operating system calls for the specified lock.
  */
  if( eFileLock==SHARED_LOCK ){
    assert( pLock->nShared==0 );
    assert( pLock->eFileLock==0 );

    /* Now get the read-lock */
    lock.l_start = SHARED_FIRST;
    lock.l_len = SHARED_SIZE;
    if( (s = fcntl(pFile->h, F_SETLK, &lock))==(-1) ){
      tErrno = errno;
    }
................................................................................
    if( s==(-1) ){
      rc = sqliteErrorFromPosixError(tErrno, SQLITE_IOERR_LOCK);
      if( IS_LOCK_ERROR(rc) ){
        pFile->lastErrno = tErrno;
      }
    }else{
      pFile->eFileLock = SHARED_LOCK;
      pFile->pOpen->nLock++;
      pLock->nShared = 1;
    }
  }else if( eFileLock==EXCLUSIVE_LOCK && pLock->nShared>1 ){
    /* We are trying for an exclusive lock but another thread in this
    ** same process is still holding a shared lock. */
    rc = SQLITE_BUSY;
  }else{
    /* The request was for a RESERVED or EXCLUSIVE lock.  It is
    ** assumed that there is a SHARED or greater lock on the file
    ** already.
................................................................................
    pFile->inNormalWrite = 1;
  }
#endif


  if( rc==SQLITE_OK ){
    pFile->eFileLock = eFileLock;
    pLock->eFileLock = eFileLock;
  }else if( eFileLock==EXCLUSIVE_LOCK ){
    pFile->eFileLock = PENDING_LOCK;
    pLock->eFileLock = PENDING_LOCK;
  }

end_lock:
  unixLeaveMutex();
  OSTRACE(("LOCK    %d %s %s (unix)\n", pFile->h, azFileLock(eFileLock), 
      rc==SQLITE_OK ? "ok" : "failed"));
  return rc;
}

/*
** Close all file descriptors accumuated in the unixOpenCnt->pUnused list.
** If all such file descriptors are closed without error, the list is
** cleared and SQLITE_OK returned.
**
** Otherwise, if an error occurs, then successfully closed file descriptor
** entries are removed from the list, and SQLITE_IOERR_CLOSE returned. 
** not deleted and SQLITE_IOERR_CLOSE returned.
*/ 
static int closePendingFds(unixFile *pFile){
  int rc = SQLITE_OK;
  struct unixOpenCnt *pOpen = pFile->pOpen;
  UnixUnusedFd *pError = 0;
  UnixUnusedFd *p;
  UnixUnusedFd *pNext;
  for(p=pOpen->pUnused; p; p=pNext){
    pNext = p->pNext;
    if( close(p->fd) ){
      pFile->lastErrno = errno;
      rc = SQLITE_IOERR_CLOSE;
      p->pNext = pError;
      pError = p;
    }else{
      sqlite3_free(p);
    }
  }
  pOpen->pUnused = pError;
  return rc;
}

/*
** Add the file descriptor used by file handle pFile to the corresponding
** pUnused list.
*/
static void setPendingFd(unixFile *pFile){
  struct unixOpenCnt *pOpen = pFile->pOpen;
  UnixUnusedFd *p = pFile->pUnused;
  p->pNext = pOpen->pUnused;
  pOpen->pUnused = p;
  pFile->h = -1;
  pFile->pUnused = 0;
}

/*
** Lower the locking level on file descriptor pFile to eFileLock.  eFileLock
** must be either NO_LOCK or SHARED_LOCK.
................................................................................
** the byte range is divided into 2 parts and the first part is unlocked then
** set to a read lock, then the other part is simply unlocked.  This works 
** around a bug in BSD NFS lockd (also seen on MacOSX 10.3+) that fails to 
** remove the write lock on a region when a read lock is set.
*/
static int _posixUnlock(sqlite3_file *id, int eFileLock, int handleNFSUnlock){
  unixFile *pFile = (unixFile*)id;
  struct unixLockInfo *pLock;
  struct flock lock;
  int rc = SQLITE_OK;
  int h;
  int tErrno;                      /* Error code from system call errors */

  assert( pFile );
  OSTRACE(("UNLOCK  %d %d was %d(%d,%d) pid=%d (unix)\n", pFile->h, eFileLock,
      pFile->eFileLock, pFile->pLock->eFileLock, pFile->pLock->nShared,
      getpid()));

  assert( eFileLock<=SHARED_LOCK );
  if( pFile->eFileLock<=eFileLock ){
    return SQLITE_OK;
  }
  if( CHECK_THREADID(pFile) ){
    return SQLITE_MISUSE_BKPT;
  }
  unixEnterMutex();
  h = pFile->h;
  pLock = pFile->pLock;
  assert( pLock->nShared!=0 );
  if( pFile->eFileLock>SHARED_LOCK ){
    assert( pLock->eFileLock==pFile->eFileLock );
    SimulateIOErrorBenign(1);
    SimulateIOError( h=(-1) )
    SimulateIOErrorBenign(0);

#ifndef NDEBUG
    /* When reducing a lock such that other processes can start
    ** reading the database file again, make sure that the
................................................................................
      }
    }
    lock.l_type = F_UNLCK;
    lock.l_whence = SEEK_SET;
    lock.l_start = PENDING_BYTE;
    lock.l_len = 2L;  assert( PENDING_BYTE+1==RESERVED_BYTE );
    if( fcntl(h, F_SETLK, &lock)!=(-1) ){
      pLock->eFileLock = SHARED_LOCK;
    }else{
      tErrno = errno;
      rc = sqliteErrorFromPosixError(tErrno, SQLITE_IOERR_UNLOCK);
      if( IS_LOCK_ERROR(rc) ){
        pFile->lastErrno = tErrno;
      }
      goto end_unlock;
    }
  }
  if( eFileLock==NO_LOCK ){
    struct unixOpenCnt *pOpen;

    /* Decrement the shared lock counter.  Release the lock using an
    ** OS call only when all threads in this same process have released
    ** the lock.
    */
    pLock->nShared--;
    if( pLock->nShared==0 ){
      lock.l_type = F_UNLCK;
      lock.l_whence = SEEK_SET;
      lock.l_start = lock.l_len = 0L;
      SimulateIOErrorBenign(1);
      SimulateIOError( h=(-1) )
      SimulateIOErrorBenign(0);
      if( fcntl(h, F_SETLK, &lock)!=(-1) ){
        pLock->eFileLock = NO_LOCK;
      }else{
        tErrno = errno;
        rc = sqliteErrorFromPosixError(tErrno, SQLITE_IOERR_UNLOCK);
        if( IS_LOCK_ERROR(rc) ){
          pFile->lastErrno = tErrno;
        }
        pLock->eFileLock = NO_LOCK;
        pFile->eFileLock = NO_LOCK;
      }
    }

    /* Decrement the count of locks against this same file.  When the
    ** count reaches zero, close any other file descriptors whose close
    ** was deferred because of outstanding locks.
    */
    pOpen = pFile->pOpen;
    pOpen->nLock--;
    assert( pOpen->nLock>=0 );
    if( pOpen->nLock==0 ){
      int rc2 = closePendingFds(pFile);
      if( rc==SQLITE_OK ){
        rc = rc2;
      }
    }
  }
	
................................................................................
*/
static int unixClose(sqlite3_file *id){
  int rc = SQLITE_OK;
  if( id ){
    unixFile *pFile = (unixFile *)id;
    unixUnlock(id, NO_LOCK);
    unixEnterMutex();
    if( pFile->pOpen && pFile->pOpen->nLock ){
      /* If there are outstanding locks, do not actually close the file just
      ** yet because that would clear those locks.  Instead, add the file
      ** descriptor to pOpen->pUnused list.  It will be automatically closed 
      ** when the last lock is cleared.
      */
      setPendingFd(pFile);
    }
    releaseLockInfo(pFile->pLock);
    releaseOpenCnt(pFile->pOpen);
    rc = closeUnixFile(id);
    unixLeaveMutex();
  }
  return rc;
}

/************** End of the posix advisory lock implementation *****************
................................................................................
  /* Check if a thread in this process holds such a lock */
  if( pFile->eFileLock>SHARED_LOCK ){
    reserved = 1;
  }
  
  /* Otherwise see if some other process holds it. */
  if( !reserved ){
    sem_t *pSem = pFile->pOpen->pSem;
    struct stat statBuf;

    if( sem_trywait(pSem)==-1 ){
      int tErrno = errno;
      if( EAGAIN != tErrno ){
        rc = sqliteErrorFromPosixError(tErrno, SQLITE_IOERR_CHECKRESERVEDLOCK);
        pFile->lastErrno = tErrno;
................................................................................
**
** This routine will only increase a lock.  Use the sqlite3OsUnlock()
** routine to lower a locking level.
*/
static int semLock(sqlite3_file *id, int eFileLock) {
  unixFile *pFile = (unixFile*)id;
  int fd;
  sem_t *pSem = pFile->pOpen->pSem;
  int rc = SQLITE_OK;

  /* if we already have a lock, it is exclusive.  
  ** Just adjust level and punt on outta here. */
  if (pFile->eFileLock > NO_LOCK) {
    pFile->eFileLock = eFileLock;
    rc = SQLITE_OK;
................................................................................
** must be either NO_LOCK or SHARED_LOCK.
**
** If the locking level of the file descriptor is already at or below
** the requested locking level, this routine is a no-op.
*/
static int semUnlock(sqlite3_file *id, int eFileLock) {
  unixFile *pFile = (unixFile*)id;
  sem_t *pSem = pFile->pOpen->pSem;

  assert( pFile );
  assert( pSem );
  OSTRACE(("UNLOCK  %d %d was %d pid=%d (sem)\n", pFile->h, eFileLock,
	   pFile->eFileLock, getpid()));
  assert( eFileLock<=SHARED_LOCK );
  
................................................................................
 */
static int semClose(sqlite3_file *id) {
  if( id ){
    unixFile *pFile = (unixFile*)id;
    semUnlock(id, NO_LOCK);
    assert( pFile );
    unixEnterMutex();
    releaseLockInfo(pFile->pLock);
    releaseOpenCnt(pFile->pOpen);
    unixLeaveMutex();
    closeUnixFile(id);
  }
  return SQLITE_OK;
}

#endif /* OS_VXWORKS */
................................................................................
  
  assert( pFile );
  afpLockingContext *context = (afpLockingContext *) pFile->lockingContext;
  if( context->reserved ){
    *pResOut = 1;
    return SQLITE_OK;
  }
  unixEnterMutex(); /* Because pFile->pLock is shared across threads */
  
  /* Check if a thread in this process holds such a lock */
  if( pFile->pLock->eFileLock>SHARED_LOCK ){
    reserved = 1;
  }
  
  /* Otherwise see if some other process holds it.
   */
  if( !reserved ){
    /* lock the RESERVED byte */
................................................................................
**
** This routine will only increase a lock.  Use the sqlite3OsUnlock()
** routine to lower a locking level.
*/
static int afpLock(sqlite3_file *id, int eFileLock){
  int rc = SQLITE_OK;
  unixFile *pFile = (unixFile*)id;
  struct unixLockInfo *pLock = pFile->pLock;
  afpLockingContext *context = (afpLockingContext *) pFile->lockingContext;
  
  assert( pFile );
  OSTRACE(("LOCK    %d %s was %s(%s,%d) pid=%d (afp)\n", pFile->h,
           azFileLock(eFileLock), azFileLock(pFile->eFileLock),
           azFileLock(pLock->eFileLock), pLock->nShared , getpid()));

  /* If there is already a lock of this type or more restrictive on the
  ** unixFile, do nothing. Don't use the afp_end_lock: exit path, as
  ** unixEnterMutex() hasn't been called yet.
  */
  if( pFile->eFileLock>=eFileLock ){
    OSTRACE(("LOCK    %d %s ok (already held) (afp)\n", pFile->h,
................................................................................
  **  (2) SQLite never explicitly requests a pendig lock.
  **  (3) A shared lock is always held when a reserve lock is requested.
  */
  assert( pFile->eFileLock!=NO_LOCK || eFileLock==SHARED_LOCK );
  assert( eFileLock!=PENDING_LOCK );
  assert( eFileLock!=RESERVED_LOCK || pFile->eFileLock==SHARED_LOCK );
  
  /* This mutex is needed because pFile->pLock is shared across threads
  */
  unixEnterMutex();

  /* Make sure the current thread owns the pFile.
  */
  rc = transferOwnership(pFile);
  if( rc!=SQLITE_OK ){
    unixLeaveMutex();
    return rc;
  }
  pLock = pFile->pLock;

  /* If some thread using this PID has a lock via a different unixFile*
  ** handle that precludes the requested lock, return BUSY.
  */
  if( (pFile->eFileLock!=pLock->eFileLock && 
       (pLock->eFileLock>=PENDING_LOCK || eFileLock>SHARED_LOCK))
     ){
    rc = SQLITE_BUSY;
    goto afp_end_lock;
  }
  
  /* If a SHARED lock is requested, and some thread using this PID already
  ** has a SHARED or RESERVED lock, then increment reference counts and
  ** return SQLITE_OK.
  */
  if( eFileLock==SHARED_LOCK && 
     (pLock->eFileLock==SHARED_LOCK || pLock->eFileLock==RESERVED_LOCK) ){
    assert( eFileLock==SHARED_LOCK );
    assert( pFile->eFileLock==0 );
    assert( pLock->nShared>0 );
    pFile->eFileLock = SHARED_LOCK;
    pLock->nShared++;
    pFile->pOpen->nLock++;
    goto afp_end_lock;
  }
    
  /* A PENDING lock is needed before acquiring a SHARED lock and before
  ** acquiring an EXCLUSIVE lock.  For the SHARED lock, the PENDING will
  ** be released.
  */
................................................................................
  /* If control gets to this point, then actually go ahead and make
  ** operating system calls for the specified lock.
  */
  if( eFileLock==SHARED_LOCK ){
    int lrc1, lrc2, lrc1Errno;
    long lk, mask;
    
    assert( pLock->nShared==0 );
    assert( pLock->eFileLock==0 );
        
    mask = (sizeof(long)==8) ? LARGEST_INT64 : 0x7fffffff;
    /* Now get the read-lock SHARED_LOCK */
    /* note that the quality of the randomness doesn't matter that much */
    lk = random(); 
    pLock->sharedByte = (lk & mask)%(SHARED_SIZE - 1);
    lrc1 = afpSetLock(context->dbPath, pFile, 
          SHARED_FIRST+pLock->sharedByte, 1, 1);
    if( IS_LOCK_ERROR(lrc1) ){
      lrc1Errno = pFile->lastErrno;
    }
    /* Drop the temporary PENDING lock */
    lrc2 = afpSetLock(context->dbPath, pFile, PENDING_BYTE, 1, 0);
    
    if( IS_LOCK_ERROR(lrc1) ) {
................................................................................
    } else if( IS_LOCK_ERROR(lrc2) ){
      rc = lrc2;
      goto afp_end_lock;
    } else if( lrc1 != SQLITE_OK ) {
      rc = lrc1;
    } else {
      pFile->eFileLock = SHARED_LOCK;
      pFile->pOpen->nLock++;
      pLock->nShared = 1;
    }
  }else if( eFileLock==EXCLUSIVE_LOCK && pLock->nShared>1 ){
    /* We are trying for an exclusive lock but another thread in this
     ** same process is still holding a shared lock. */
    rc = SQLITE_BUSY;
  }else{
    /* The request was for a RESERVED or EXCLUSIVE lock.  It is
    ** assumed that there is a SHARED or greater lock on the file
    ** already.
................................................................................
    if (!failed && eFileLock == EXCLUSIVE_LOCK) {
      /* Acquire an EXCLUSIVE lock */
        
      /* Remove the shared lock before trying the range.  we'll need to 
      ** reestablish the shared lock if we can't get the  afpUnlock
      */
      if( !(failed = afpSetLock(context->dbPath, pFile, SHARED_FIRST +
                         pLock->sharedByte, 1, 0)) ){
        int failed2 = SQLITE_OK;
        /* now attemmpt to get the exclusive lock range */
        failed = afpSetLock(context->dbPath, pFile, SHARED_FIRST, 
                               SHARED_SIZE, 1);
        if( failed && (failed2 = afpSetLock(context->dbPath, pFile, 
                       SHARED_FIRST + pLock->sharedByte, 1, 1)) ){
          /* Can't reestablish the shared lock.  Sqlite can't deal, this is
          ** a critical I/O error
          */
          rc = ((failed & SQLITE_IOERR) == SQLITE_IOERR) ? failed2 : 
               SQLITE_IOERR_LOCK;
          goto afp_end_lock;
        } 
................................................................................
    if( failed ){
      rc = failed;
    }
  }
  
  if( rc==SQLITE_OK ){
    pFile->eFileLock = eFileLock;
    pLock->eFileLock = eFileLock;
  }else if( eFileLock==EXCLUSIVE_LOCK ){
    pFile->eFileLock = PENDING_LOCK;
    pLock->eFileLock = PENDING_LOCK;
  }
  
afp_end_lock:
  unixLeaveMutex();
  OSTRACE(("LOCK    %d %s %s (afp)\n", pFile->h, azFileLock(eFileLock), 
         rc==SQLITE_OK ? "ok" : "failed"));
  return rc;
................................................................................
**
** If the locking level of the file descriptor is already at or below
** the requested locking level, this routine is a no-op.
*/
static int afpUnlock(sqlite3_file *id, int eFileLock) {
  int rc = SQLITE_OK;
  unixFile *pFile = (unixFile*)id;
  struct unixLockInfo *pLock;
  afpLockingContext *context = (afpLockingContext *) pFile->lockingContext;
  int skipShared = 0;
#ifdef SQLITE_TEST
  int h = pFile->h;
#endif

  assert( pFile );
  OSTRACE(("UNLOCK  %d %d was %d(%d,%d) pid=%d (afp)\n", pFile->h, eFileLock,
           pFile->eFileLock, pFile->pLock->eFileLock, pFile->pLock->nShared,
           getpid()));

  assert( eFileLock<=SHARED_LOCK );
  if( pFile->eFileLock<=eFileLock ){
    return SQLITE_OK;
  }
  if( CHECK_THREADID(pFile) ){
    return SQLITE_MISUSE_BKPT;
  }
  unixEnterMutex();
  pLock = pFile->pLock;
  assert( pLock->nShared!=0 );
  if( pFile->eFileLock>SHARED_LOCK ){
    assert( pLock->eFileLock==pFile->eFileLock );
    SimulateIOErrorBenign(1);
    SimulateIOError( h=(-1) )
    SimulateIOErrorBenign(0);
    
#ifndef NDEBUG
    /* When reducing a lock such that other processes can start
    ** reading the database file again, make sure that the
................................................................................
           || pFile->dbUpdate==0
           || pFile->transCntrChng==1 );
    pFile->inNormalWrite = 0;
#endif
    
    if( pFile->eFileLock==EXCLUSIVE_LOCK ){
      rc = afpSetLock(context->dbPath, pFile, SHARED_FIRST, SHARED_SIZE, 0);
      if( rc==SQLITE_OK && (eFileLock==SHARED_LOCK || pLock->nShared>1) ){
        /* only re-establish the shared lock if necessary */
        int sharedLockByte = SHARED_FIRST+pLock->sharedByte;
        rc = afpSetLock(context->dbPath, pFile, sharedLockByte, 1, 1);
      } else {
        skipShared = 1;
      }
    }
    if( rc==SQLITE_OK && pFile->eFileLock>=PENDING_LOCK ){
      rc = afpSetLock(context->dbPath, pFile, PENDING_BYTE, 1, 0);
................................................................................
    } 
    if( rc==SQLITE_OK && pFile->eFileLock>=RESERVED_LOCK && context->reserved ){
      rc = afpSetLock(context->dbPath, pFile, RESERVED_BYTE, 1, 0);
      if( !rc ){ 
        context->reserved = 0; 
      }
    }
    if( rc==SQLITE_OK && (eFileLock==SHARED_LOCK || pLock->nShared>1)){
      pLock->eFileLock = SHARED_LOCK;
    }
  }
  if( rc==SQLITE_OK && eFileLock==NO_LOCK ){

    /* Decrement the shared lock counter.  Release the lock using an
    ** OS call only when all threads in this same process have released
    ** the lock.
    */
    unsigned long long sharedLockByte = SHARED_FIRST+pLock->sharedByte;
    pLock->nShared--;
    if( pLock->nShared==0 ){
      SimulateIOErrorBenign(1);
      SimulateIOError( h=(-1) )
      SimulateIOErrorBenign(0);
      if( !skipShared ){
        rc = afpSetLock(context->dbPath, pFile, sharedLockByte, 1, 0);
      }
      if( !rc ){
        pLock->eFileLock = NO_LOCK;
        pFile->eFileLock = NO_LOCK;
      }
    }
    if( rc==SQLITE_OK ){
      struct unixOpenCnt *pOpen = pFile->pOpen;
        
      pOpen->nLock--;
      assert( pOpen->nLock>=0 );
      if( pOpen->nLock==0 ){
        rc = closePendingFds(pFile);
      }
    }
  }
  
  unixLeaveMutex();
  if( rc==SQLITE_OK ) pFile->eFileLock = eFileLock;
................................................................................
*/
static int afpClose(sqlite3_file *id) {
  int rc = SQLITE_OK;
  if( id ){
    unixFile *pFile = (unixFile*)id;
    afpUnlock(id, NO_LOCK);
    unixEnterMutex();
    if( pFile->pOpen && pFile->pOpen->nLock ){
      /* If there are outstanding locks, do not actually close the file just
      ** yet because that would clear those locks.  Instead, add the file
      ** descriptor to pOpen->aPending.  It will be automatically closed when
      ** the last lock is cleared.
      */
      setPendingFd(pFile);
    }
    releaseLockInfo(pFile->pLock);
    releaseOpenCnt(pFile->pOpen);
    sqlite3_free(pFile->lockingContext);
    rc = closeUnixFile(id);
    unixLeaveMutex();
  }
  return rc;
}

................................................................................
  SimulateIOError( rc=1 );
  if( rc!=0 ){
    ((unixFile*)id)->lastErrno = errno;
    return SQLITE_IOERR_FSTAT;
  }
  *pSize = buf.st_size;

  /* When opening a zero-size database, the findLockInfo() procedure
  ** writes a single byte into that file in order to work around a bug
  ** in the OS-X msdos filesystem.  In order to avoid problems with upper
  ** layers, we need to report this file size as zero even though it is
  ** really 1.   Ticket #3260.
  */
  if( *pSize==1 ) *pSize = 0;

................................................................................
  int noLock,             /* Omit locking if true */
  int isDelete            /* Delete on close if true */
){
  const sqlite3_io_methods *pLockingStyle;
  unixFile *pNew = (unixFile *)pId;
  int rc = SQLITE_OK;

  assert( pNew->pLock==NULL );
  assert( pNew->pOpen==NULL );

  /* Parameter isDelete is only used on vxworks. Express this explicitly 
  ** here to prevent compiler warnings about unused parameters.
  */
  UNUSED_PARAMETER(isDelete);

  OSTRACE(("OPEN    %-3d %s\n", h, zFilename));
  pNew->h = h;
  pNew->dirfd = dirfd;
  SET_THREADID(pNew);
  pNew->fileFlags = 0;
  assert( zFilename==0 || zFilename[0]=='/' );  /* Never a relative pathname */
  pNew->zPath = zFilename;

#if OS_VXWORKS
  pNew->pId = vxworksFindFileId(zFilename);
  if( pNew->pId==0 ){
................................................................................

  if( pLockingStyle == &posixIoMethods
#if defined(__APPLE__) && SQLITE_ENABLE_LOCKING_STYLE
    || pLockingStyle == &nfsIoMethods
#endif
  ){
    unixEnterMutex();
    rc = findLockInfo(pNew, &pNew->pLock, &pNew->pOpen);
    if( rc!=SQLITE_OK ){
      /* If an error occured in findLockInfo(), close the file descriptor
      ** immediately, before releasing the mutex. findLockInfo() may fail
      ** in two scenarios:
      **
      **   (a) A call to fstat() failed.
      **   (b) A malloc failed.
      **
      ** Scenario (b) may only occur if the process is holding no other
      ** file descriptors open on the same file. If there were other file
      ** descriptors on this file, then no malloc would be required by
      ** findLockInfo(). If this is the case, it is quite safe to close
      ** handle h - as it is guaranteed that no posix locks will be released
      ** by doing so.
      **
      ** If scenario (a) caused the error then things are not so safe. The
      ** implicit assumption here is that if fstat() fails, things are in
      ** such bad shape that dropping a lock or two doesn't matter much.
      */
................................................................................
      /* NB: zFilename exists and remains valid until the file is closed
      ** according to requirement F11141.  So we do not need to make a
      ** copy of the filename. */
      pCtx->dbPath = zFilename;
      pCtx->reserved = 0;
      srandomdev();
      unixEnterMutex();
      rc = findLockInfo(pNew, &pNew->pLock, &pNew->pOpen);
      if( rc!=SQLITE_OK ){
        sqlite3_free(pNew->lockingContext);
        close(h);
        h = -1;
      }
      unixLeaveMutex();        
    }
................................................................................

#if OS_VXWORKS
  else if( pLockingStyle == &semIoMethods ){
    /* Named semaphore locking uses the file path so it needs to be
    ** included in the semLockingContext
    */
    unixEnterMutex();
    rc = findLockInfo(pNew, &pNew->pLock, &pNew->pOpen);
    if( (rc==SQLITE_OK) && (pNew->pOpen->pSem==NULL) ){
      char *zSemName = pNew->pOpen->aSemName;
      int n;
      sqlite3_snprintf(MAX_PATHNAME, zSemName, "/%s.sem",
                       pNew->pId->zCanonicalName);
      for( n=1; zSemName[n]; n++ )
        if( zSemName[n]=='/' ) zSemName[n] = '_';
      pNew->pOpen->pSem = sem_open(zSemName, O_CREAT, 0666, 1);
      if( pNew->pOpen->pSem == SEM_FAILED ){
        rc = SQLITE_NOMEM;
        pNew->pOpen->aSemName[0] = '\0';
      }
    }
    unixLeaveMutex();
  }
#endif
  
  pNew->lastErrno = 0;
................................................................................
  ** For this reason, if an error occurs in the stat() call here, it is
  ** ignored and -1 is returned. The caller will try to open a new file
  ** descriptor on the same path, fail, and return an error to SQLite.
  **
  ** Even if a subsequent open() call does succeed, the consequences of
  ** not searching for a resusable file descriptor are not dire.  */
  if( 0==stat(zPath, &sStat) ){
    struct unixOpenCnt *pOpen;

    unixEnterMutex();
    pOpen = openList;
    while( pOpen && (pOpen->fileId.dev!=sStat.st_dev
                     || pOpen->fileId.ino!=sStat.st_ino) ){
       pOpen = pOpen->pNext;
    }
    if( pOpen ){
      UnixUnusedFd **pp;
      for(pp=&pOpen->pUnused; *pp && (*pp)->flags!=flags; pp=&((*pp)->pNext));
      pUnused = *pp;
      if( pUnused ){
        *pp = pUnused->pNext;
      }
    }
    unixLeaveMutex();
  }
................................................................................
      
      /* update conch with host and path (this will fail if other process
      ** has a shared lock already), if the host id matches, use the big
      ** stick.
      */
      futimes(conchFile->h, NULL);
      if( hostIdMatch && !createConch ){
        if( conchFile->pLock && conchFile->pLock->nShared>1 ){
          /* We are trying for an exclusive lock but another thread in this
           ** same process is still holding a shared lock. */
          rc = SQLITE_BUSY;
        } else {          
          rc = proxyConchLock(pFile, myHostID, EXCLUSIVE_LOCK);
        }
      }else{

/*
** The unixFile structure is subclass of sqlite3_file specific to the unix
** VFS implementations.
*/
typedef struct unixFile unixFile;
struct unixFile {
  sqlite3_io_methods const *pMethod;  /* Always the first entry */

  struct unixInodeInfo *pInode;       /* Info about locks on this inode */
  int h;                              /* The file descriptor */
  int dirfd;                          /* File descriptor for the directory */
  unsigned char eFileLock;            /* The type of lock held on this fd */
  int lastErrno;                      /* The unix errno from last I/O error */
  void *lockingContext;               /* Locking style specific state */
  UnixUnusedFd *pUnused;              /* Pre-allocated UnixUnusedFd */
  int fileFlags;                      /* Miscellanous flags */
  const char *zPath;                  /* Name of the file */
  unixShm *pShm;                      /* Shared memory segment information */
#if SQLITE_ENABLE_LOCKING_STYLE
  int openFlags;                      /* The flags specified at open() */
#endif
#if SQLITE_ENABLE_LOCKING_STYLE || defined(__APPLE__)
  unsigned fsFlags;                   /* cached details from statfs() */
#endif



#if OS_VXWORKS
  int isDelete;                       /* Delete on close if true */
  struct vxworksFileId *pId;          /* Unique file ID */
#endif
#ifndef NDEBUG
  /* The next group of variables are used to track whether or not the
  ** transaction counter in bytes 24-27 of database files are updated
  ** whenever any part of the database changes.  An assertion fault will
  ** occur if a file is updated without also updating the transaction
  ** counter.  This test is made to avoid new problems similar to the
................................................................................
#else
#define threadid 0
#endif


/*
** Helper functions to obtain and relinquish the global mutex. The
** global mutex is used to protect the unixInodeInfo and
** vxworksFileId objects used by this file, all of which may be 
** shared by multiple threads.
**
** Function unixMutexHeld() is used to assert() that the global mutex 
** is held when required. This function is only used as part of assert() 
** statements. e.g.
**
................................................................................
** POSIX lock if the internal lock structure transitions between
** a locked and an unlocked state.
**
** But wait:  there are yet more problems with POSIX advisory locks.
**
** If you close a file descriptor that points to a file that has locks,
** all locks on that file that are owned by the current process are
** released.  To work around this problem, each unixInodeInfo object
** maintains a count of the number of pending locks on tha inode.

** When an attempt is made to close an unixFile, if there are
** other unixFile open on the same inode that are holding locks, the call
** to close() the file descriptor is deferred until all of the locks clear.
** The unixInodeInfo structure keeps a list of file descriptors that need to
** be closed and that list is walked (and cleared) when the last lock
** clears.
**
** Yet another problem:  LinuxThreads do not play well with posix locks.
**
** Many older versions of linux use the LinuxThreads library which is
** not posix compliant.  Under LinuxThreads, a lock created by thread
................................................................................
** A cannot be modified or overridden by a different thread B.
** Only thread A can modify the lock.  Locking behavior is correct
** if the appliation uses the newer Native Posix Thread Library (NPTL)
** on linux - with NPTL a lock created by thread A can override locks
** in thread B.  But there is no way to know at compile-time which
** threading library is being used.  So there is no way to know at
** compile-time whether or not thread A can override locks on thread B.
** One has to do a run-time check to discover the behavior of the
** current process.
**
** SQLite used to support LinuxThreads.  But support for LinuxThreads
** was dropped beginning with version 3.7.0.  SQLite will still work with
** LinuxThreads provided that (1) there is no more than one connection 
** per database file in the same process and (2) database connections
** do not move across threads.


*/

/*
























** An instance of the following structure serves as the key used
** to locate a particular unixInodeInfo object.

*/
struct unixFileId {
  dev_t dev;                  /* Device number */
#if OS_VXWORKS
  struct vxworksFileId *pId;  /* Unique file ID for vxworks. */
#else
  ino_t ino;                  /* Inode number */
#endif
};


















/*
** An instance of the following structure is allocated for each open
** inode.  Or, on LinuxThreads, there is one of these structures for
** each inode opened by each thread.
**
** A single inode can have multiple file descriptors, so each unixFile
** structure contains a pointer to an instance of this object and this
** object keeps a count of the number of unixFile pointing to it.
*/
struct unixInodeInfo {
  struct unixFileId fileId;       /* The lookup key */
  int nShared;                    /* Number of SHARED locks held */
  int eFileLock;                  /* One of SHARED_LOCK, RESERVED_LOCK etc. */
  int nRef;                       /* Number of pointers to this structure */
#if defined(SQLITE_ENABLE_LOCKING_STYLE)
  unsigned long long sharedByte;  /* for AFP simulated shared lock */
#endif



















  int nLock;                      /* Number of outstanding file locks */
  UnixUnusedFd *pUnused;          /* Unused file descriptors to close */
#if OS_VXWORKS
  sem_t *pSem;                    /* Named POSIX semaphore */
  char aSemName[MAX_PATHNAME+2];  /* Name of that semaphore */
#endif

  struct unixInodeInfo *pNext;    /* List of all unixInodeInfo objects */
  struct unixInodeInfo *pPrev;    /*    .... doubly linked */
};

/*




** A lists of all unixInodeInfo objects.
*/

static struct unixInodeInfo *inodeList = 0;

/*




























































































** Release a unixInodeInfo structure previously allocated by findInodeInfo().
**
** The mutex entered using the unixEnterMutex() function must be held
** when this function is called.
*/

static void releaseInodeInfo(struct unixInodeInfo *pInode){
  assert( unixMutexHeld() );
  if( pInode ){
    pInode->nRef--;
    if( pInode->nRef==0 ){
      if( pInode->pPrev ){
        assert( pInode->pPrev->pNext==pInode );
        pInode->pPrev->pNext = pInode->pNext;
      }else{


        assert( inodeList==pInode );
        inodeList = pInode->pNext;
      }
      if( pInode->pNext ){
        assert( pInode->pNext->pPrev==pInode );
        pInode->pNext->pPrev = pInode->pPrev;
      }
      sqlite3_free(pInode);
    }
  }
}

/*






































** Given a file descriptor, locate the unixInodeInfo object that
** describes that file descriptor.  Create a new one if necessary.  The
** return value might be uninitialized if an error occurs.
**
** The mutex entered using the unixEnterMutex() function must be held
** when this function is called.
**
** Return an appropriate error code.
*/
static int findInodeInfo(
  unixFile *pFile,               /* Unix file with file desc used in the key */


  struct unixInodeInfo **ppInode /* Return the unixInodeInfo object here */
){
  int rc;                        /* System call return code */
  int fd;                        /* The file descriptor for pFile */

  struct unixFileId fileId;          /* Lookup key for the unixInodeInfo */
  struct stat statbuf;               /* Low-level file information */


  struct unixInodeInfo *pInode = 0;  /* Candidate unixInodeInfo object */

  assert( unixMutexHeld() );

  /* Get low-level information about the file that we can used to
  ** create a unique name for the file.
  */
  fd = pFile->h;
................................................................................
    if( rc!=0 ){
      pFile->lastErrno = errno;
      return SQLITE_IOERR;
    }
  }
#endif

  memset(&fileId, 0, sizeof(fileId));
  fileId.dev = statbuf.st_dev;
#if OS_VXWORKS
  fileId.pId = pFile->pId;
#else
  fileId.ino = statbuf.st_ino;
#endif
  pInode = inodeList;



































  while( pInode && memcmp(&fileId, &pInode->fileId, sizeof(fileId)) ){
    pInode = pInode->pNext;
  }

  if( pInode==0 ){
    pInode = sqlite3_malloc( sizeof(*pInode) );


    if( pInode==0 ){
      return SQLITE_NOMEM;

    }
    memset(pInode, 0, sizeof(*pInode));

    memcpy(&pInode->fileId, &fileId, sizeof(fileId));
    pInode->nRef = 1;
    pInode->pNext = inodeList;

    pInode->pPrev = 0;
    if( inodeList ) inodeList->pPrev = pInode;
    inodeList = pInode;
  }else{
    pInode->nRef++;
  }
























  *ppInode = pInode;
  return SQLITE_OK;
}






























/*
** This routine checks if there is a RESERVED lock held on the specified
** file by this or any other process. If such a lock is held, set *pResOut
** to a non-zero value otherwise *pResOut is set to zero.  The return value
** is set to SQLITE_OK unless an I/O error occurs during lock checking.
................................................................................
  int rc = SQLITE_OK;
  int reserved = 0;
  unixFile *pFile = (unixFile*)id;

  SimulateIOError( return SQLITE_IOERR_CHECKRESERVEDLOCK; );

  assert( pFile );
  unixEnterMutex(); /* Because pFile->pInode is shared across threads */

  /* Check if a thread in this process holds such a lock */
  if( pFile->pInode->eFileLock>SHARED_LOCK ){
    reserved = 1;
  }

  /* Otherwise see if some other process holds it.
  */
#ifndef __DJGPP__
  if( !reserved ){
................................................................................
  ** The reason a single byte cannot be used instead of the 'shared byte
  ** range' is that some versions of windows do not support read-locks. By
  ** locking a random byte from a range, concurrent SHARED locks may exist
  ** even if the locking primitive used is always a write-lock.
  */
  int rc = SQLITE_OK;
  unixFile *pFile = (unixFile*)id;
  struct unixInodeInfo *pInode = pFile->pInode;
  struct flock lock;
  int s = 0;
  int tErrno = 0;

  assert( pFile );
  OSTRACE(("LOCK    %d %s was %s(%s,%d) pid=%d (unix)\n", pFile->h,
      azFileLock(eFileLock), azFileLock(pFile->eFileLock),
      azFileLock(pInode->eFileLock), pInode->nShared , getpid()));

  /* If there is already a lock of this type or more restrictive on the
  ** unixFile, do nothing. Don't use the end_lock: exit path, as
  ** unixEnterMutex() hasn't been called yet.
  */
  if( pFile->eFileLock>=eFileLock ){
    OSTRACE(("LOCK    %d %s ok (already held) (unix)\n", pFile->h,
................................................................................
  **  (2) SQLite never explicitly requests a pendig lock.
  **  (3) A shared lock is always held when a reserve lock is requested.
  */
  assert( pFile->eFileLock!=NO_LOCK || eFileLock==SHARED_LOCK );
  assert( eFileLock!=PENDING_LOCK );
  assert( eFileLock!=RESERVED_LOCK || pFile->eFileLock==SHARED_LOCK );

  /* This mutex is needed because pFile->pInode is shared across threads
  */
  unixEnterMutex();








  pInode = pFile->pInode;

  /* If some thread using this PID has a lock via a different unixFile*
  ** handle that precludes the requested lock, return BUSY.
  */
  if( (pFile->eFileLock!=pInode->eFileLock && 
          (pInode->eFileLock>=PENDING_LOCK || eFileLock>SHARED_LOCK))
  ){
    rc = SQLITE_BUSY;
    goto end_lock;
  }

  /* If a SHARED lock is requested, and some thread using this PID already
  ** has a SHARED or RESERVED lock, then increment reference counts and
  ** return SQLITE_OK.
  */
  if( eFileLock==SHARED_LOCK && 
      (pInode->eFileLock==SHARED_LOCK || pInode->eFileLock==RESERVED_LOCK) ){
    assert( eFileLock==SHARED_LOCK );
    assert( pFile->eFileLock==0 );
    assert( pInode->nShared>0 );
    pFile->eFileLock = SHARED_LOCK;
    pInode->nShared++;
    pInode->nLock++;
    goto end_lock;
  }


  /* A PENDING lock is needed before acquiring a SHARED lock and before
  ** acquiring an EXCLUSIVE lock.  For the SHARED lock, the PENDING will
  ** be released.
................................................................................
  }


  /* If control gets to this point, then actually go ahead and make
  ** operating system calls for the specified lock.
  */
  if( eFileLock==SHARED_LOCK ){
    assert( pInode->nShared==0 );
    assert( pInode->eFileLock==0 );

    /* Now get the read-lock */
    lock.l_start = SHARED_FIRST;
    lock.l_len = SHARED_SIZE;
    if( (s = fcntl(pFile->h, F_SETLK, &lock))==(-1) ){
      tErrno = errno;
    }
................................................................................
    if( s==(-1) ){
      rc = sqliteErrorFromPosixError(tErrno, SQLITE_IOERR_LOCK);
      if( IS_LOCK_ERROR(rc) ){
        pFile->lastErrno = tErrno;
      }
    }else{
      pFile->eFileLock = SHARED_LOCK;
      pInode->nLock++;
      pInode->nShared = 1;
    }
  }else if( eFileLock==EXCLUSIVE_LOCK && pInode->nShared>1 ){
    /* We are trying for an exclusive lock but another thread in this
    ** same process is still holding a shared lock. */
    rc = SQLITE_BUSY;
  }else{
    /* The request was for a RESERVED or EXCLUSIVE lock.  It is
    ** assumed that there is a SHARED or greater lock on the file
    ** already.
................................................................................
    pFile->inNormalWrite = 1;
  }
#endif


  if( rc==SQLITE_OK ){
    pFile->eFileLock = eFileLock;
    pInode->eFileLock = eFileLock;
  }else if( eFileLock==EXCLUSIVE_LOCK ){
    pFile->eFileLock = PENDING_LOCK;
    pInode->eFileLock = PENDING_LOCK;
  }

end_lock:
  unixLeaveMutex();
  OSTRACE(("LOCK    %d %s %s (unix)\n", pFile->h, azFileLock(eFileLock), 
      rc==SQLITE_OK ? "ok" : "failed"));
  return rc;
}

/*
** Close all file descriptors accumuated in the unixInodeInfo->pUnused list.
** If all such file descriptors are closed without error, the list is
** cleared and SQLITE_OK returned.
**
** Otherwise, if an error occurs, then successfully closed file descriptor
** entries are removed from the list, and SQLITE_IOERR_CLOSE returned. 
** not deleted and SQLITE_IOERR_CLOSE returned.
*/ 
static int closePendingFds(unixFile *pFile){
  int rc = SQLITE_OK;
  struct unixInodeInfo *pInode = pFile->pInode;
  UnixUnusedFd *pError = 0;
  UnixUnusedFd *p;
  UnixUnusedFd *pNext;
  for(p=pInode->pUnused; p; p=pNext){
    pNext = p->pNext;
    if( close(p->fd) ){
      pFile->lastErrno = errno;
      rc = SQLITE_IOERR_CLOSE;
      p->pNext = pError;
      pError = p;
    }else{
      sqlite3_free(p);
    }
  }
  pInode->pUnused = pError;
  return rc;
}

/*
** Add the file descriptor used by file handle pFile to the corresponding
** pUnused list.
*/
static void setPendingFd(unixFile *pFile){
  struct unixInodeInfo *pInode = pFile->pInode;
  UnixUnusedFd *p = pFile->pUnused;
  p->pNext = pInode->pUnused;
  pInode->pUnused = p;
  pFile->h = -1;
  pFile->pUnused = 0;
}

/*
** Lower the locking level on file descriptor pFile to eFileLock.  eFileLock
** must be either NO_LOCK or SHARED_LOCK.
................................................................................
** the byte range is divided into 2 parts and the first part is unlocked then
** set to a read lock, then the other part is simply unlocked.  This works 
** around a bug in BSD NFS lockd (also seen on MacOSX 10.3+) that fails to 
** remove the write lock on a region when a read lock is set.
*/
static int _posixUnlock(sqlite3_file *id, int eFileLock, int handleNFSUnlock){
  unixFile *pFile = (unixFile*)id;
  struct unixInodeInfo *pInode;
  struct flock lock;
  int rc = SQLITE_OK;
  int h;
  int tErrno;                      /* Error code from system call errors */

  assert( pFile );
  OSTRACE(("UNLOCK  %d %d was %d(%d,%d) pid=%d (unix)\n", pFile->h, eFileLock,
      pFile->eFileLock, pFile->pInode->eFileLock, pFile->pInode->nShared,
      getpid()));

  assert( eFileLock<=SHARED_LOCK );
  if( pFile->eFileLock<=eFileLock ){
    return SQLITE_OK;
  }



  unixEnterMutex();
  h = pFile->h;
  pInode = pFile->pInode;
  assert( pInode->nShared!=0 );
  if( pFile->eFileLock>SHARED_LOCK ){
    assert( pInode->eFileLock==pFile->eFileLock );
    SimulateIOErrorBenign(1);
    SimulateIOError( h=(-1) )
    SimulateIOErrorBenign(0);

#ifndef NDEBUG
    /* When reducing a lock such that other processes can start
    ** reading the database file again, make sure that the
................................................................................
      }
    }
    lock.l_type = F_UNLCK;
    lock.l_whence = SEEK_SET;
    lock.l_start = PENDING_BYTE;
    lock.l_len = 2L;  assert( PENDING_BYTE+1==RESERVED_BYTE );
    if( fcntl(h, F_SETLK, &lock)!=(-1) ){
      pInode->eFileLock = SHARED_LOCK;
    }else{
      tErrno = errno;
      rc = sqliteErrorFromPosixError(tErrno, SQLITE_IOERR_UNLOCK);
      if( IS_LOCK_ERROR(rc) ){
        pFile->lastErrno = tErrno;
      }
      goto end_unlock;
    }
  }
  if( eFileLock==NO_LOCK ){


    /* Decrement the shared lock counter.  Release the lock using an
    ** OS call only when all threads in this same process have released
    ** the lock.
    */
    pInode->nShared--;
    if( pInode->nShared==0 ){
      lock.l_type = F_UNLCK;
      lock.l_whence = SEEK_SET;
      lock.l_start = lock.l_len = 0L;
      SimulateIOErrorBenign(1);
      SimulateIOError( h=(-1) )
      SimulateIOErrorBenign(0);
      if( fcntl(h, F_SETLK, &lock)!=(-1) ){
        pInode->eFileLock = NO_LOCK;
      }else{
        tErrno = errno;
        rc = sqliteErrorFromPosixError(tErrno, SQLITE_IOERR_UNLOCK);
        if( IS_LOCK_ERROR(rc) ){
          pFile->lastErrno = tErrno;
        }
        pInode->eFileLock = NO_LOCK;
        pFile->eFileLock = NO_LOCK;
      }
    }

    /* Decrement the count of locks against this same file.  When the
    ** count reaches zero, close any other file descriptors whose close
    ** was deferred because of outstanding locks.
    */

    pInode->nLock--;
    assert( pInode->nLock>=0 );
    if( pInode->nLock==0 ){
      int rc2 = closePendingFds(pFile);
      if( rc==SQLITE_OK ){
        rc = rc2;
      }
    }
  }
	
................................................................................
*/
static int unixClose(sqlite3_file *id){
  int rc = SQLITE_OK;
  if( id ){
    unixFile *pFile = (unixFile *)id;
    unixUnlock(id, NO_LOCK);
    unixEnterMutex();
    if( pFile->pInode && pFile->pInode->nLock ){
      /* If there are outstanding locks, do not actually close the file just
      ** yet because that would clear those locks.  Instead, add the file
      ** descriptor to pInode->pUnused list.  It will be automatically closed 
      ** when the last lock is cleared.
      */
      setPendingFd(pFile);
    }
    releaseInodeInfo(pFile->pInode);

    rc = closeUnixFile(id);
    unixLeaveMutex();
  }
  return rc;
}

/************** End of the posix advisory lock implementation *****************
................................................................................
  /* Check if a thread in this process holds such a lock */
  if( pFile->eFileLock>SHARED_LOCK ){
    reserved = 1;
  }
  
  /* Otherwise see if some other process holds it. */
  if( !reserved ){
    sem_t *pSem = pFile->pInode->pSem;
    struct stat statBuf;

    if( sem_trywait(pSem)==-1 ){
      int tErrno = errno;
      if( EAGAIN != tErrno ){
        rc = sqliteErrorFromPosixError(tErrno, SQLITE_IOERR_CHECKRESERVEDLOCK);
        pFile->lastErrno = tErrno;
................................................................................
**
** This routine will only increase a lock.  Use the sqlite3OsUnlock()
** routine to lower a locking level.
*/
static int semLock(sqlite3_file *id, int eFileLock) {
  unixFile *pFile = (unixFile*)id;
  int fd;
  sem_t *pSem = pFile->pInode->pSem;
  int rc = SQLITE_OK;

  /* if we already have a lock, it is exclusive.  
  ** Just adjust level and punt on outta here. */
  if (pFile->eFileLock > NO_LOCK) {
    pFile->eFileLock = eFileLock;
    rc = SQLITE_OK;
................................................................................
** must be either NO_LOCK or SHARED_LOCK.
**
** If the locking level of the file descriptor is already at or below
** the requested locking level, this routine is a no-op.
*/
static int semUnlock(sqlite3_file *id, int eFileLock) {
  unixFile *pFile = (unixFile*)id;
  sem_t *pSem = pFile->pInode->pSem;

  assert( pFile );
  assert( pSem );
  OSTRACE(("UNLOCK  %d %d was %d pid=%d (sem)\n", pFile->h, eFileLock,
	   pFile->eFileLock, getpid()));
  assert( eFileLock<=SHARED_LOCK );
  
................................................................................
 */
static int semClose(sqlite3_file *id) {
  if( id ){
    unixFile *pFile = (unixFile*)id;
    semUnlock(id, NO_LOCK);
    assert( pFile );
    unixEnterMutex();
    releaseLockInfo(pFile->pInode);

    unixLeaveMutex();
    closeUnixFile(id);
  }
  return SQLITE_OK;
}

#endif /* OS_VXWORKS */
................................................................................
  
  assert( pFile );
  afpLockingContext *context = (afpLockingContext *) pFile->lockingContext;
  if( context->reserved ){
    *pResOut = 1;
    return SQLITE_OK;
  }
  unixEnterMutex(); /* Because pFile->pInode is shared across threads */
  
  /* Check if a thread in this process holds such a lock */
  if( pFile->pInode->eFileLock>SHARED_LOCK ){
    reserved = 1;
  }
  
  /* Otherwise see if some other process holds it.
   */
  if( !reserved ){
    /* lock the RESERVED byte */
................................................................................
**
** This routine will only increase a lock.  Use the sqlite3OsUnlock()
** routine to lower a locking level.
*/
static int afpLock(sqlite3_file *id, int eFileLock){
  int rc = SQLITE_OK;
  unixFile *pFile = (unixFile*)id;
  struct unixInodeInfo *pInode = pFile->pInode;
  afpLockingContext *context = (afpLockingContext *) pFile->lockingContext;
  
  assert( pFile );
  OSTRACE(("LOCK    %d %s was %s(%s,%d) pid=%d (afp)\n", pFile->h,
           azFileLock(eFileLock), azFileLock(pFile->eFileLock),
           azFileLock(pInode->eFileLock), pInode->nShared , getpid()));

  /* If there is already a lock of this type or more restrictive on the
  ** unixFile, do nothing. Don't use the afp_end_lock: exit path, as
  ** unixEnterMutex() hasn't been called yet.
  */
  if( pFile->eFileLock>=eFileLock ){
    OSTRACE(("LOCK    %d %s ok (already held) (afp)\n", pFile->h,
................................................................................
  **  (2) SQLite never explicitly requests a pendig lock.
  **  (3) A shared lock is always held when a reserve lock is requested.
  */
  assert( pFile->eFileLock!=NO_LOCK || eFileLock==SHARED_LOCK );
  assert( eFileLock!=PENDING_LOCK );
  assert( eFileLock!=RESERVED_LOCK || pFile->eFileLock==SHARED_LOCK );
  
  /* This mutex is needed because pFile->pInode is shared across threads
  */
  unixEnterMutex();








  pInode = pFile->pInode;

  /* If some thread using this PID has a lock via a different unixFile*
  ** handle that precludes the requested lock, return BUSY.
  */
  if( (pFile->eFileLock!=pInode->eFileLock && 
       (pInode->eFileLock>=PENDING_LOCK || eFileLock>SHARED_LOCK))
     ){
    rc = SQLITE_BUSY;
    goto afp_end_lock;
  }
  
  /* If a SHARED lock is requested, and some thread using this PID already
  ** has a SHARED or RESERVED lock, then increment reference counts and
  ** return SQLITE_OK.
  */
  if( eFileLock==SHARED_LOCK && 
     (pInode->eFileLock==SHARED_LOCK || pInode->eFileLock==RESERVED_LOCK) ){
    assert( eFileLock==SHARED_LOCK );
    assert( pFile->eFileLock==0 );
    assert( pInode->nShared>0 );
    pFile->eFileLock = SHARED_LOCK;
    pInode->nShared++;
    pInode->nLock++;
    goto afp_end_lock;
  }
    
  /* A PENDING lock is needed before acquiring a SHARED lock and before
  ** acquiring an EXCLUSIVE lock.  For the SHARED lock, the PENDING will
  ** be released.
  */
................................................................................
  /* If control gets to this point, then actually go ahead and make
  ** operating system calls for the specified lock.
  */
  if( eFileLock==SHARED_LOCK ){
    int lrc1, lrc2, lrc1Errno;
    long lk, mask;
    
    assert( pInode->nShared==0 );
    assert( pInode->eFileLock==0 );
        
    mask = (sizeof(long)==8) ? LARGEST_INT64 : 0x7fffffff;
    /* Now get the read-lock SHARED_LOCK */
    /* note that the quality of the randomness doesn't matter that much */
    lk = random(); 
    pInode->sharedByte = (lk & mask)%(SHARED_SIZE - 1);
    lrc1 = afpSetLock(context->dbPath, pFile, 
          SHARED_FIRST+pInode->sharedByte, 1, 1);
    if( IS_LOCK_ERROR(lrc1) ){
      lrc1Errno = pFile->lastErrno;
    }
    /* Drop the temporary PENDING lock */
    lrc2 = afpSetLock(context->dbPath, pFile, PENDING_BYTE, 1, 0);
    
    if( IS_LOCK_ERROR(lrc1) ) {
................................................................................
    } else if( IS_LOCK_ERROR(lrc2) ){
      rc = lrc2;
      goto afp_end_lock;
    } else if( lrc1 != SQLITE_OK ) {
      rc = lrc1;
    } else {
      pFile->eFileLock = SHARED_LOCK;
      pInode->nLock++;
      pInode->nShared = 1;
    }
  }else if( eFileLock==EXCLUSIVE_LOCK && pInode->nShared>1 ){
    /* We are trying for an exclusive lock but another thread in this
     ** same process is still holding a shared lock. */
    rc = SQLITE_BUSY;
  }else{
    /* The request was for a RESERVED or EXCLUSIVE lock.  It is
    ** assumed that there is a SHARED or greater lock on the file
    ** already.
................................................................................
    if (!failed && eFileLock == EXCLUSIVE_LOCK) {
      /* Acquire an EXCLUSIVE lock */
        
      /* Remove the shared lock before trying the range.  we'll need to 
      ** reestablish the shared lock if we can't get the  afpUnlock
      */
      if( !(failed = afpSetLock(context->dbPath, pFile, SHARED_FIRST +
                         pInode->sharedByte, 1, 0)) ){
        int failed2 = SQLITE_OK;
        /* now attemmpt to get the exclusive lock range */
        failed = afpSetLock(context->dbPath, pFile, SHARED_FIRST, 
                               SHARED_SIZE, 1);
        if( failed && (failed2 = afpSetLock(context->dbPath, pFile, 
                       SHARED_FIRST + pInode->sharedByte, 1, 1)) ){
          /* Can't reestablish the shared lock.  Sqlite can't deal, this is
          ** a critical I/O error
          */
          rc = ((failed & SQLITE_IOERR) == SQLITE_IOERR) ? failed2 : 
               SQLITE_IOERR_LOCK;
          goto afp_end_lock;
        } 
................................................................................
    if( failed ){
      rc = failed;
    }
  }
  
  if( rc==SQLITE_OK ){
    pFile->eFileLock = eFileLock;
    pInode->eFileLock = eFileLock;
  }else if( eFileLock==EXCLUSIVE_LOCK ){
    pFile->eFileLock = PENDING_LOCK;
    pInode->eFileLock = PENDING_LOCK;
  }
  
afp_end_lock:
  unixLeaveMutex();
  OSTRACE(("LOCK    %d %s %s (afp)\n", pFile->h, azFileLock(eFileLock), 
         rc==SQLITE_OK ? "ok" : "failed"));
  return rc;
................................................................................
**
** If the locking level of the file descriptor is already at or below
** the requested locking level, this routine is a no-op.
*/
static int afpUnlock(sqlite3_file *id, int eFileLock) {
  int rc = SQLITE_OK;
  unixFile *pFile = (unixFile*)id;
  struct unixInodeInfo *pInode;
  afpLockingContext *context = (afpLockingContext *) pFile->lockingContext;
  int skipShared = 0;
#ifdef SQLITE_TEST
  int h = pFile->h;
#endif

  assert( pFile );
  OSTRACE(("UNLOCK  %d %d was %d(%d,%d) pid=%d (afp)\n", pFile->h, eFileLock,
           pFile->eFileLock, pFile->pInode->eFileLock, pFile->pInode->nShared,
           getpid()));

  assert( eFileLock<=SHARED_LOCK );
  if( pFile->eFileLock<=eFileLock ){
    return SQLITE_OK;
  }



  unixEnterMutex();
  pInode = pFile->pInode;
  assert( pInode->nShared!=0 );
  if( pFile->eFileLock>SHARED_LOCK ){
    assert( pInode->eFileLock==pFile->eFileLock );
    SimulateIOErrorBenign(1);
    SimulateIOError( h=(-1) )
    SimulateIOErrorBenign(0);
    
#ifndef NDEBUG
    /* When reducing a lock such that other processes can start
    ** reading the database file again, make sure that the
................................................................................
           || pFile->dbUpdate==0
           || pFile->transCntrChng==1 );
    pFile->inNormalWrite = 0;
#endif
    
    if( pFile->eFileLock==EXCLUSIVE_LOCK ){
      rc = afpSetLock(context->dbPath, pFile, SHARED_FIRST, SHARED_SIZE, 0);
      if( rc==SQLITE_OK && (eFileLock==SHARED_LOCK || pInode->nShared>1) ){
        /* only re-establish the shared lock if necessary */
        int sharedLockByte = SHARED_FIRST+pInode->sharedByte;
        rc = afpSetLock(context->dbPath, pFile, sharedLockByte, 1, 1);
      } else {
        skipShared = 1;
      }
    }
    if( rc==SQLITE_OK && pFile->eFileLock>=PENDING_LOCK ){
      rc = afpSetLock(context->dbPath, pFile, PENDING_BYTE, 1, 0);
................................................................................
    } 
    if( rc==SQLITE_OK && pFile->eFileLock>=RESERVED_LOCK && context->reserved ){
      rc = afpSetLock(context->dbPath, pFile, RESERVED_BYTE, 1, 0);
      if( !rc ){ 
        context->reserved = 0; 
      }
    }
    if( rc==SQLITE_OK && (eFileLock==SHARED_LOCK || pInode->nShared>1)){
      pInode->eFileLock = SHARED_LOCK;
    }
  }
  if( rc==SQLITE_OK && eFileLock==NO_LOCK ){

    /* Decrement the shared lock counter.  Release the lock using an
    ** OS call only when all threads in this same process have released
    ** the lock.
    */
    unsigned long long sharedLockByte = SHARED_FIRST+pInode->sharedByte;
    pInode->nShared--;
    if( pInode->nShared==0 ){
      SimulateIOErrorBenign(1);
      SimulateIOError( h=(-1) )
      SimulateIOErrorBenign(0);
      if( !skipShared ){
        rc = afpSetLock(context->dbPath, pFile, sharedLockByte, 1, 0);
      }
      if( !rc ){
        pInode->eFileLock = NO_LOCK;
        pFile->eFileLock = NO_LOCK;
      }
    }
    if( rc==SQLITE_OK ){


      pInode->nLock--;
      assert( pInode->nLock>=0 );
      if( pInode->nLock==0 ){
        rc = closePendingFds(pFile);
      }
    }
  }
  
  unixLeaveMutex();
  if( rc==SQLITE_OK ) pFile->eFileLock = eFileLock;
................................................................................
*/
static int afpClose(sqlite3_file *id) {
  int rc = SQLITE_OK;
  if( id ){
    unixFile *pFile = (unixFile*)id;
    afpUnlock(id, NO_LOCK);
    unixEnterMutex();
    if( pFile->pInode && pFile->pInode->nLock ){
      /* If there are outstanding locks, do not actually close the file just
      ** yet because that would clear those locks.  Instead, add the file
      ** descriptor to pInode->aPending.  It will be automatically closed when
      ** the last lock is cleared.
      */
      setPendingFd(pFile);
    }
    releaseLockInfo(pFile->pInode);

    sqlite3_free(pFile->lockingContext);
    rc = closeUnixFile(id);
    unixLeaveMutex();
  }
  return rc;
}

................................................................................
  SimulateIOError( rc=1 );
  if( rc!=0 ){
    ((unixFile*)id)->lastErrno = errno;
    return SQLITE_IOERR_FSTAT;
  }
  *pSize = buf.st_size;

  /* When opening a zero-size database, the findInodeInfo() procedure
  ** writes a single byte into that file in order to work around a bug
  ** in the OS-X msdos filesystem.  In order to avoid problems with upper
  ** layers, we need to report this file size as zero even though it is
  ** really 1.   Ticket #3260.
  */
  if( *pSize==1 ) *pSize = 0;

................................................................................
  int noLock,             /* Omit locking if true */
  int isDelete            /* Delete on close if true */
){
  const sqlite3_io_methods *pLockingStyle;
  unixFile *pNew = (unixFile *)pId;
  int rc = SQLITE_OK;

  assert( pNew->pInode==NULL );


  /* Parameter isDelete is only used on vxworks. Express this explicitly 
  ** here to prevent compiler warnings about unused parameters.
  */
  UNUSED_PARAMETER(isDelete);

  OSTRACE(("OPEN    %-3d %s\n", h, zFilename));
  pNew->h = h;
  pNew->dirfd = dirfd;

  pNew->fileFlags = 0;
  assert( zFilename==0 || zFilename[0]=='/' );  /* Never a relative pathname */
  pNew->zPath = zFilename;

#if OS_VXWORKS
  pNew->pId = vxworksFindFileId(zFilename);
  if( pNew->pId==0 ){
................................................................................

  if( pLockingStyle == &posixIoMethods
#if defined(__APPLE__) && SQLITE_ENABLE_LOCKING_STYLE
    || pLockingStyle == &nfsIoMethods
#endif
  ){
    unixEnterMutex();
    rc = findInodeInfo(pNew, &pNew->pInode);
    if( rc!=SQLITE_OK ){
      /* If an error occured in findInodeInfo(), close the file descriptor
      ** immediately, before releasing the mutex. findInodeInfo() may fail
      ** in two scenarios:
      **
      **   (a) A call to fstat() failed.
      **   (b) A malloc failed.
      **
      ** Scenario (b) may only occur if the process is holding no other
      ** file descriptors open on the same file. If there were other file
      ** descriptors on this file, then no malloc would be required by
      ** findInodeInfo(). If this is the case, it is quite safe to close
      ** handle h - as it is guaranteed that no posix locks will be released
      ** by doing so.
      **
      ** If scenario (a) caused the error then things are not so safe. The
      ** implicit assumption here is that if fstat() fails, things are in
      ** such bad shape that dropping a lock or two doesn't matter much.
      */
................................................................................
      /* NB: zFilename exists and remains valid until the file is closed
      ** according to requirement F11141.  So we do not need to make a
      ** copy of the filename. */
      pCtx->dbPath = zFilename;
      pCtx->reserved = 0;
      srandomdev();
      unixEnterMutex();
      rc = findInodeInfo(pNew, &pNew->pInode);
      if( rc!=SQLITE_OK ){
        sqlite3_free(pNew->lockingContext);
        close(h);
        h = -1;
      }
      unixLeaveMutex();        
    }
................................................................................

#if OS_VXWORKS
  else if( pLockingStyle == &semIoMethods ){
    /* Named semaphore locking uses the file path so it needs to be
    ** included in the semLockingContext
    */
    unixEnterMutex();
    rc = findInodeInfo(pNew, &pNew->pInode);
    if( (rc==SQLITE_OK) && (pNew->pInode->pSem==NULL) ){
      char *zSemName = pNew->pInode->aSemName;
      int n;
      sqlite3_snprintf(MAX_PATHNAME, zSemName, "/%s.sem",
                       pNew->pId->zCanonicalName);
      for( n=1; zSemName[n]; n++ )
        if( zSemName[n]=='/' ) zSemName[n] = '_';
      pNew->pInode->pSem = sem_open(zSemName, O_CREAT, 0666, 1);
      if( pNew->pInode->pSem == SEM_FAILED ){
        rc = SQLITE_NOMEM;
        pNew->pInode->aSemName[0] = '\0';
      }
    }
    unixLeaveMutex();
  }
#endif
  
  pNew->lastErrno = 0;
................................................................................
  ** For this reason, if an error occurs in the stat() call here, it is
  ** ignored and -1 is returned. The caller will try to open a new file
  ** descriptor on the same path, fail, and return an error to SQLite.
  **
  ** Even if a subsequent open() call does succeed, the consequences of
  ** not searching for a resusable file descriptor are not dire.  */
  if( 0==stat(zPath, &sStat) ){
    struct unixInodeInfo *pInode;

    unixEnterMutex();
    pInode = inodeList;
    while( pInode && (pInode->fileId.dev!=sStat.st_dev
                     || pInode->fileId.ino!=sStat.st_ino) ){
       pInode = pInode->pNext;
    }
    if( pInode ){
      UnixUnusedFd **pp;
      for(pp=&pInode->pUnused; *pp && (*pp)->flags!=flags; pp=&((*pp)->pNext));
      pUnused = *pp;
      if( pUnused ){
        *pp = pUnused->pNext;
      }
    }
    unixLeaveMutex();
  }
................................................................................
      
      /* update conch with host and path (this will fail if other process
      ** has a shared lock already), if the host id matches, use the big
      ** stick.
      */
      futimes(conchFile->h, NULL);
      if( hostIdMatch && !createConch ){
        if( conchFile->pInode && conchFile->pInode->nShared>1 ){
          /* We are trying for an exclusive lock but another thread in this
           ** same process is still holding a shared lock. */
          rc = SQLITE_BUSY;
        } else {          
          rc = proxyConchLock(pFile, myHostID, EXCLUSIVE_LOCK);
        }
      }else{

  extern int sqlite3_sync_count, sqlite3_fullsync_count;
  extern int sqlite3_opentemp_count;
  extern int sqlite3_like_count;
  extern int sqlite3_xferopt_count;
  extern int sqlite3_pager_readdb_count;
  extern int sqlite3_pager_writedb_count;
  extern int sqlite3_pager_writej_count;
#if defined(__linux__) && defined(SQLITE_TEST) && SQLITE_THREADSAFE
  extern int threadsOverrideEachOthersLocks;
#endif
#if SQLITE_OS_WIN
  extern int sqlite3_os_type;
#endif
#ifdef SQLITE_DEBUG
  extern int sqlite3WhereTrace;
  extern int sqlite3OSTrace;
  extern int sqlite3VdbeAddopTrace;
................................................................................
      (char*)&sqlite3_pager_writedb_count, TCL_LINK_INT);
  Tcl_LinkVar(interp, "sqlite3_pager_writej_count",
      (char*)&sqlite3_pager_writej_count, TCL_LINK_INT);
#ifndef SQLITE_OMIT_UTF16
  Tcl_LinkVar(interp, "unaligned_string_counter",
      (char*)&unaligned_string_counter, TCL_LINK_INT);
#endif
#if defined(__linux__) && defined(SQLITE_TEST) && SQLITE_THREADSAFE
  Tcl_LinkVar(interp, "threadsOverrideEachOthersLocks",
      (char*)&threadsOverrideEachOthersLocks, TCL_LINK_INT);
#endif
#ifndef SQLITE_OMIT_UTF16
  Tcl_LinkVar(interp, "sqlite_last_needed_collation",
      (char*)&pzNeededCollation, TCL_LINK_STRING|TCL_LINK_READ_ONLY);
#endif
#if SQLITE_OS_WIN
  Tcl_LinkVar(interp, "sqlite_os_type",
      (char*)&sqlite3_os_type, TCL_LINK_INT);

  extern int sqlite3_sync_count, sqlite3_fullsync_count;
  extern int sqlite3_opentemp_count;
  extern int sqlite3_like_count;
  extern int sqlite3_xferopt_count;
  extern int sqlite3_pager_readdb_count;
  extern int sqlite3_pager_writedb_count;
  extern int sqlite3_pager_writej_count;



#if SQLITE_OS_WIN
  extern int sqlite3_os_type;
#endif
#ifdef SQLITE_DEBUG
  extern int sqlite3WhereTrace;
  extern int sqlite3OSTrace;
  extern int sqlite3VdbeAddopTrace;
................................................................................
      (char*)&sqlite3_pager_writedb_count, TCL_LINK_INT);
  Tcl_LinkVar(interp, "sqlite3_pager_writej_count",
      (char*)&sqlite3_pager_writej_count, TCL_LINK_INT);
#ifndef SQLITE_OMIT_UTF16
  Tcl_LinkVar(interp, "unaligned_string_counter",
      (char*)&unaligned_string_counter, TCL_LINK_INT);
#endif




#ifndef SQLITE_OMIT_UTF16
  Tcl_LinkVar(interp, "sqlite_last_needed_collation",
      (char*)&pzNeededCollation, TCL_LINK_STRING|TCL_LINK_READ_ONLY);
#endif
#if SQLITE_OS_WIN
  Tcl_LinkVar(interp, "sqlite_os_type",
      (char*)&sqlite3_os_type, TCL_LINK_INT);

# Skip this whole file if the thread testing code is not enabled
#
if {[llength [info command thread_step]]==0 || [sqlite3 -has-codec]} {
  finish_test
  return
}
if {![info exists threadsOverrideEachOthersLocks]} {
  finish_test
  return
}

# Create some data to work with
#
do_test thread1-1.1 {
  execsql {
    CREATE TABLE t1(a,b);
    INSERT INTO t1 VALUES(1,'abcdefgh');
................................................................................
do_test thread2-2.9 {
  thread_finalize B
  thread_result B
} {SQLITE_OK}
thread_halt A
thread_halt B

# Save the original (correct) value of threadsOverrideEachOthersLocks
# so that it can be restored.  If this value is left set incorrectly, lots
# of things will go wrong in future tests.
#
set orig_threadOverride $threadsOverrideEachOthersLocks

# Pretend we are on a system (like RedHat9) were threads do not
# override each others locks.
#
set threadsOverrideEachOthersLocks 0

# Verify that we can move database connections between threads as
# long as no locks are held.
#
do_test thread2-3.1 {
  thread_create A test.db
  set DB [thread_db_get A]
  thread_halt A
} {}
do_test thread2-3.2 {
  set STMT [sqlite3_prepare $DB {SELECT a FROM t1 LIMIT 1} -1 TAIL]
  sqlite3_step $STMT
} SQLITE_ROW
do_test thread2-3.3 {
  sqlite3_column_int $STMT 0
} 1
do_test thread2-3.4 {
  sqlite3_finalize $STMT
} SQLITE_OK
do_test thread2-3.5 {
  set STMT [sqlite3_prepare $DB {SELECT max(a) FROM t1} -1 TAIL]
  sqlite3_step $STMT
} SQLITE_ROW
do_test thread2-3.6 {
  sqlite3_column_int $STMT 0
} 8
do_test thread2-3.7 {
  sqlite3_finalize $STMT
} SQLITE_OK
do_test thread2-3.8 {
  sqlite3_close $DB
} {SQLITE_OK}

do_test thread2-3.10 {
  thread_create A test.db
  thread_compile A {SELECT a FROM t1 LIMIT 1}
  thread_step A
  thread_finalize A
  set DB [thread_db_get A]
  thread_halt A
} {}
do_test thread2-3.11 {
  set STMT [sqlite3_prepare $DB {SELECT a FROM t1 LIMIT 1} -1 TAIL]
  sqlite3_step $STMT
} SQLITE_ROW
do_test thread2-3.12 {
  sqlite3_column_int $STMT 0
} 1
do_test thread2-3.13 {
  sqlite3_finalize $STMT
} SQLITE_OK
do_test thread2-3.14 {
  sqlite3_close $DB
} SQLITE_OK

do_test thread2-3.20 {
  thread_create A test.db
  thread_compile A {SELECT a FROM t1 LIMIT 3}
  thread_step A
  set STMT [thread_stmt_get A]
  set DB [thread_db_get A]
  sqlite3_step $STMT
} SQLITE_ROW
do_test thread2-3.22 {
  sqlite3_column_int $STMT 0
} 2
do_test thread2-3.23 {
  # The unlock fails here.  But because we never check the return
  # code from sqlite3OsUnlock (because we cannot do anything about it
  # if it fails) we do not realize that an error has occurred.
  breakpoint
  sqlite3_finalize $STMT
} SQLITE_OK
do_test thread2-3.25 {
  thread_db_put A $DB
  thread_halt A
} {}

do_test thread2-3.30 {
  thread_create A test.db
  thread_compile A {BEGIN}
  thread_step A
  thread_finalize A
  thread_compile A {SELECT a FROM t1 LIMIT 1}
  thread_step A
  thread_finalize A
  set DB [thread_db_get A]
  set STMT [sqlite3_prepare $DB {INSERT INTO t1 VALUES(99,'error')} -1 TAIL]
  sqlite3_step $STMT
} SQLITE_ERROR
do_test thread2-3.32 {
  sqlite3_finalize $STMT
} SQLITE_MISUSE
do_test thread2-3.33 {
  thread_db_put A $DB
  thread_halt A
} {}

# VERY important to set the override flag back to its true value.
#
set threadsOverrideEachOthersLocks $orig_threadOverride

# Also important to halt the worker threads, which are using spin
# locks and eating away CPU cycles.
#
thread_halt *   
finish_test

# Skip this whole file if the thread testing code is not enabled
#
if {[llength [info command thread_step]]==0 || [sqlite3 -has-codec]} {
  finish_test
  return
}





# Create some data to work with
#
do_test thread1-1.1 {
  execsql {
    CREATE TABLE t1(a,b);
    INSERT INTO t1 VALUES(1,'abcdefgh');
................................................................................
do_test thread2-2.9 {
  thread_finalize B
  thread_result B
} {SQLITE_OK}
thread_halt A
thread_halt B

















































































































# Also important to halt the worker threads, which are using spin
# locks and eating away CPU cycles.
#
thread_halt *   
finish_test

# 2008 November 11
#
# The author disclaims copyright to this source code.  In place of
# a legal notice, here is a blessing:
#
#    May you do good and not evil.
#    May you find forgiveness for yourself and forgive others.
#    May you share freely, never taking more than you give.
#
#***********************************************************************
#     
# $Id: tkt3472.test,v 1.4 2008/12/03 22:32:45 drh Exp $

set testdir [file dirname $argv0]
source $testdir/tester.tcl

if {![info exists threadsOverrideEachOthersLocks]} {
  finish_test
  return
}

set ::correctvalue $threadsOverrideEachOthersLocks
puts "threadsOverrideEachOthersLocks = $::correctvalue"

do_test tkt3472-1.1 {
  db close
  set threadsOverrideEachOthersLocks -1
  sqlite3 db test.db
  set threadsOverrideEachOthersLocks
} $::correctvalue

do_test tkt3472-1.2 {
  db close
  set threadsOverrideEachOthersLocks -1
  sqlite3 db test.db -readonly 1
  set threadsOverrideEachOthersLocks
} $::correctvalue

finish_test